Integr. equ. oper. theory 45 (2003) 437–460 0378-620X/040437-24 c 2003 Birkh¨ auser Verlag Basel/Switzerland
Integral Equations and Operator Theory
H−n–perturbations of Self-adjoint Operators and Krein’s Resolvent Formula Pavel Kurasov To my father
Abstract. Supersingular H−n rank one perturbations of an arbitrary positive self-adjoint operator A acting in the Hilbert space H are investigated. The operator corresponding to the formal expression Aα = A + αϕ, ·ϕ, α ∈ R, ϕ ∈ H−n (A), is determined as a regular operator with pure real spectrum acting in a certain extended Hilbert space H ⊃ H. The resolvent of the operator so defined is given by a certain generalization of Krein’s resolvent formula. It is proven that the spectral properties of the operator are described by generalized Nevanlinna functions. The results of [24] are extended to the case of arbitrary integer n ≥ 4. Mathematics Subject Classification (2000). Primary 47A55, 81Q10; Secondary 47A20, 81Q15. Keywords. Singular perturbations, Krein’s formula, Nevanlinna functions.
1. Introduction Finite rank singular perturbations of self-adjoint operators have been studied intensively during the recent years [2,3,4,6,8,15,16,20,28,29,37]. In particular partial differential operators with point interactions are described in [1,9], following the pioneering work by F.Berezin and L.Faddeev from 1961 [7]. One of the main mathematical tools to study spectral properties of these operators is the well-known Krein’s resolvent formula relating the resolvents of two self-adjoint extensions of one symmetric operator having finite or infinite deficiency indices [21,27,34]. Selfcontained presentation of this theory can be found in recent papers [4,14]. In the current paper we continue our studies of singular rank one perturbations of a positive self-adjoint operator A acting in the Hilbert space H. The
438
Kurasov
IEOT
perturbed operator can formally be defined by Aα = A + αϕ, ·ϕ,
(1.1)
where α ∈ R is a coupling constant and ϕ is the singular vector describing the interaction. To measure the singularity of the interaction one can use the scale Hs of Hilbert spaces1 associated with the self-adjoint operator A acting in the Hilbert space H Dom (A)
H
(Dom (A))∗
. . . ⊂ H4 ⊂ H3 ⊂ H2 ⊂ H1 ⊂ H0 ⊂ H−1 ⊂
⊂ H−3 ⊂ H−4 ⊂ . . . (1.2) We say that the interaction is from the class H−n if and only if ϕ ∈ H−n \ H−n+1 . The singular interactions from the classes H−1 and H−2 can be defined using operators acting in the original Hilbert space H. The perturbation term αϕ, ·ϕ is infinitesimally form bounded with respect to the operator A, if ϕ ∈ H−1 . The perturbed operator is uniquely defined using the KLMN theorem in [33]. The perturbed operator in the case of H−2 perturbations is not defined uniquely - in this case a one parameter family of self-adjoint operators corresponds to formal expression (1.1) [19,2,3,4]. The current paper is devoted to so-called supersingular perturbations defined by vectors from H−n , n ≥ 3. Such perturbations have been studied using a certain extension of the original Hilbert space. In [35,36,12,13] rank one supersingular perturbations were defined using self-adjoint operators acting in Pontryagin spaces. It was shown that the spectral properties of these models are described by generalized Nevanlinna functions with a finite number of negative squares [16]. Similar ideas were used in [18,30-32] where concrete problems of mathematical physics were attacked. Different physicists and mathematicians tried to define supersingular perturbations [10,11,17,5]. In [23] supersingular rank one perturbation of positive self-adjoint operator have been defined without any use of spaces with indefinite metrics. The approach was limited to the case of H−3 perturbations. In [24] we were able to make one step further and describe all supersingular perturbations from the class H−4 . It turns out that these perturbed operators are not self-adjoint but are regular. The following definition was introduced in [24]: H−2
Definition 1.1. Densely defined operator B is called regular if its domain coincides with the domain of the adjoint operator. The set of regular operators contains all self-adjoint operators. The class of self-adjoint operators can be characterized by one additional restriction: the densely defined operator B is self-adjoint if it is regular and symmetric. Obviously the set of regular operators extends the set of self-adjoint operators enormously. The operators corresponding to supersingular interactions have one additional remarkable property: 1 For
the precise definition, see Section 2.
Vol. 45 (2003)
Krein’s Formula for H−n –perturbations
439
The spectrum of the operator is purely real. This property makes it possible to use high order perturbations in physical applications. The formula describing the resolvent of the perturbed operator is similar to the celebrated Krein’s formula. The spectral properties of the operators are described by generalized Nevanlinna functions. The aim of the current paper is to generalize the ideas of [24] to the case of arbitrary supersingular perturbations. For the convenience of the reader we recall the necessary preliminary facts concerning singular rank one perturbations in Section2. The main ideas of the current paper are described in this section. In particular it is proposed to define the perturbed operator as a restriction of a certain maximal operator. The maximal operator and the extended Hilbert space used to construct supersingular perturbations are described in Section 3. The family of regular operators corresponding to such singular perturbation is obtained in Section 4. The resolvent formula describing supersingular perturbation is calculated. The relations between this formula and Krein’s resolvent formula are investigated.
2. Rank One Perturbations and the Extensions Theory The present paper is devoted to the construction of the operator describing rank one supersingular perturbation of a given positive self-adjoint operator A acting in a certain Hilbert space H, given formally by (1.1). Description of the recent developments in this area can be found in [2,3,15,19,22,26,37]. It has been shown that if ϕ belongs to the original Hilbert space H then the perturbation αϕ, ·ϕ is a bounded symmetric operator and the perturbed operator Aα is self-adjoint on the domain of the original operator A. The resolvent of the perturbed operator is given by 1 1 1 1 1 = − 1 ϕ. (2.1) ϕ, · 1 ¯ Aα − λ A−λ A − λ A − λ + ϕ, ϕ α A−λ All spectral properties of the perturbed operator Aα are described by the Nevan1 ϕ (See e.g. [4]). linna function Q(λ) = ϕ, A−λ Consider now the scale of Hilbert spaces Hs associated with the positive operator A. The norm in each space Hs is defined by U 2Hs = U, (A + 1)s U , where ·, · is the scalar product in the original Hilbert space H. In order to avoid misunderstanding only the scale of Hilbert spaces associated with the original operator A and the original Hilbert space H will be considered throughout the paper. All perturbations defined by vectors ϕ not from the original Hilbert space H are called singular. These perturbations are characterized by the fact that the domain of the perturbed operator does not coincide with the domain of the original one. In the case ϕ ∈ H−1 \ H the perturbation is relatively form bounded with respect to the sesquilinear form of the operator A and the perturbed operator can be determined using the form perturbation technique. The resolvent of the perturbed
440
Kurasov
IEOT
operator is again given by (2.1). The main difference is that the domain of the perturbed operator does not coincide with the domain of the original operator in general, but the perturbed operator is uniquely defined and is a self-adjoint operator acting in the original Hilbert space H [37,2]. Another way to define the perturbed operator is using the extension theory for symmetric operators. It is obvious that the perturbed and original operators coincide on the linear set of functions U satisfying the condition ϕ, U = 0.
(2.2)
Then the perturbed operator is an extension of the original operator restricted to this linear set. If ϕ ∈ H−1 \ H the restricted operator is a symmetric operator with the deficiency indices (1, 1). Its self-adjoint extension corresponding to the formal expression (1.1) is uniquely defined. The resolvent of the perturbed operator can be described using Krein’s formula [21,27], which coincides with formula (2.1) in this case. The case ϕ ∈ H−2 \ H−1 can be treated using the extension theory for symmetric operators, since the perturbation is not form bounded with respect to the original operator. The restricted symmetric operator can be defined in a way similar to H−1 -case. But the perturbed operator is not uniquely defined anymore. One can only conclude that the perturbed operator is equal to one of the self-adjoint extensions of the restricted operator. All such operators can be parametrized by one real parameter γ ∈ R ∪ {∞} as follows 1 1 1 1 1 = − ϕ. (2.3) ϕ, · 1+λ 1 ¯ Aγ − λ A − λ γ + ϕ, A−λ A − λ A − λ A+1 ϕ A relation between the real parameter γ describing the self-adjoint extensions of the restricted operator and the (additive) coupling parameter α appearing in formula (1.1) cannot be established without additional assumptions like homogeneity of the original operator and the perturbation vector.2 The Nevanlinna function 1+λ 1 Q−2 (λ) = ϕ, A−λ A+1 ϕ can be considered as a regularization of the resolvent 1 ϕ, A−λ ϕ which is not defined in the case of ϕ ∈ H−2 \ H−1 : 1 1+λ 1 1 formally ϕ = ϕ, ϕ − ϕ, ϕ. (2.4) A−λA+1 A−λ A+1 Observe that the two scalar products appearing in the right hand side of the last formula are not defined for ϕ ∈ H−2 \ H−1 , but their difference is in contrast well 1 defined. The Nevanlinna function ϕ, 1+Aλ A−λ A2 +1 ϕ just coincides with Krein’s Qfunction appearing in the formula for the difference between the resolvents of two different self-adjoint extensions of a symmetric operator with deficiency indices (1, 1) [21,27]. The next step is to consider ϕ ∈ H−3 . The restriction defined by (2.2) is defined only if one considers the original operator A as an operator acting in the Hilbert space H1 . Then the domain of the unperturbed operator A coincides with Q−2 (λ) = ϕ,
2 This
approach has been developed in [2,3].
Krein’s Formula for H−n –perturbations
Vol. 45 (2003)
441
the space H3 and the restriction (2.2) determines a symmetric operator. On the other hand formula (2.1) is valid only if one considers the extended Hilbert space 1 ϕ ∈ H−1 . It turns out that such an ension is in fact one containing vectors A−λ dimensional, since for arbitrary λ, µ ∈ C \ R+ the following inclusion is valid 1 1 1 ϕ− ϕ = (ϕ − µ) ϕ ∈ H1 . A−λ A−µ (A − λ)(A − µ) Hence it is enough to include the one dimensional subspace generated by the 1 ϕ only. Hence the perturbed operator can be defined in the Hilbert vector A+1 space H−3 = H1 ⊕ C equipped with the natural embedding ρ−3 ρ−3 :
H−3 → H−1 1 U = (U, u1 ) → U + u1 A+1 ϕ.
(2.5)
The perturbed operator corresponding to the formal expression (1.1) has been constructed in [23] by first defining a certain maximal operator acting in H−3 and then restricting it to a self-adjoint operator. The maximal operator is similar to the adjoint operator appearing in the restriction-extension procedure used to construct H−2 -perturbations. The set of self-adjoint restrictions of the maximal operator are described by one real parameter. Therefore formula (1.1) does not determine the perturbed operator uniquely, but a one parameter family of operators like in the case of H−2 -perturbations. The resolvent of the perturbed operator restricted to the original Hilbert space is given by the formula 1 1 1 1 1 |H1 = − ϕ, ϕ, · ρ−3 2 (λ+1) ¯ 1 Aθ −λ A−λ (λ + 1) cot θ + ϕ, A−λ A− λ 2 ϕ − 1 A−λ (A+1)
(2.6) where θ ∈ [0, π) is the real number parametrizing the restrictions. The similarity between formulas (2.1) and (2.6) is obvious. The function Q(λ)−3
= formally
=
ϕ,
1 (λ + 1)2 ϕ A − λ (A + 1)2
(2.7)
1 1 1 ϕ − ϕ, ϕ − (λ + 1)ϕ, ϕ, ϕ A−λ A+1 (A + 1)2
is a double regularization of the resolvent function. This function describes the spectral properties of the self-adjoint perturbed operator. Supersingular perturbation from the class H−4 have been studied in [24]. Our original aim was simply to generalize the ideas developed in [23] to the case of more singular perturbations. The main difference with the case ϕ ∈ H−3 is that the original operator A should be considered as an operator acting in the Hilbert space H2 from the scale of Hilbert spaces. Moreover this Hilbert space should be extended to include not only the vector g1 =
1 ϕ ∈ H−2 A+1
442
Kurasov
IEOT
but the vector g2 =
1 ϕ∈H (A + 1)2
as well. Hence one has to consider the Hilbert space H−4 = H2 ⊕ C2
(2.8)
equipped with the standard imbedding ρ−4 :
H−4 → H−2 1 1 U = (U, u2 , u1 ) → U + u2 (A+1) 2 ϕ + u1 A+1 ϕ.
(2.9)
The maximal operator can be defined in a way similar to H−3 -perturbations. The main difference is that any symmetric restriction of the maximal operator is not self-adjoint. Hence no self-adjoint operator corresponds to the formal expression (1.1). Instead one can consider the restrictions of the maximal operator that are regular operators. All such restrictions are parametrized by one real parameter in a way similar to H−3 perturbations. The real and imaginary parts of these operators were calculated explicitly. The resolvent of the perturbed operator was calculated as well and it was shown that the spectrum of the perturbed regular operator is pure real. The resolvent restricted to the original Hilbert space is given by a formula similar to Krein’s formula (2.3). All spectral properties of the perturbed operator are described by the Nevanlinna function Q given by Q−4 (λ)
=
ϕ,
1 (λ + 1)3 ϕ A − λ (A + 1)3
formally
ϕ,
1 1 1 ϕ − ϕ, ϕ − (λ + 1)ϕ, ϕ A−λ A+1 (A + 1)2
=
−(λ + 1)2 ϕ,
(2.10)
1 ϕ (A + 1)3
The aim of the current paper is to generalize the ideas of [24] to the case of arbitrary supersingular perturbations from the class H−n , n ≥ 4.
3. The Extended Hilbert Space and the Maximal Operator The operator corresponding to the formal expression (1.1) will be constructed as a restriction of a certain maximal operator acting in a certain extended Hilbert space. The extended Hilbert space and the maximal operator are described in the current section. To avoid a non-essential discussion we limit our consideration to the case where ϕ ∈ H−n \ H−n+1 , n ≥ 4. 3 3 The
cases n = 1, 2, 3 are well-described in the literature.
Krein’s Formula for H−n –perturbations
Vol. 45 (2003)
443
Following the ideas expressed in Section 2, we consider the Hilbert space H ≡ H−n = C(n−2) ⊕ Hn−2 4 equipped with the scalar product
U, V
=
u ¯ 1 v1 + . . . + u ¯n−3 vn−3 + u ¯n−2 vn−2 + U, V Hn−2 (n−2)
= u, v Cn−2 + U, (1 + A)
(3.1) V ,
where we used the following vector notation U = (u, U ), u = (u1 , u2 , . . . , un−2 ) V = (v , V ), u = (v1 , v2 , . . . , vn−2 ). Different scalar products can be defined in the vectors space H. The simplest case is considered in the current paper in order to avoid unnecessary complications. The general case will be studied in one of the following publications. The space H can be embedded into the space H−n+2 as follows ρU = u1 g1 + u2 g2 + . . . + un−2 gn−2 + U =
n−2
(3.2) uk gk + U,
k=1
where the vectors gk , k = 1, 2, . . . , n − 2 are defined by 1 1 gk−1 = ϕ, k = 1, 2, . . . , n − 2. g0 = ϕ, gk = A+1 (A + 1)k
(3.3)
Note that the embedding operator ρ depends on the order n. The operator A can be considered as an operator acting in the scale of Hilbert spaces. Recall that the spaces Hs ⊂ H ⊂ H−s form a Gelfand triplet for s = 1, 2, . . .: H is a Hilbert space and Hs∗ = H−s with respect to the pairing defined by the scalar product of H. Consider arbitrary Gelfand triplet K ⊂ H ⊂ K∗ . Let B be a densely defined operator in the space K then the triplet adjoint operator B † acting in K∗ is defined on the domain Dom (B † ) = {f ∈ K∗ : g ∈ Dom (B) ⇒ |Bg, f | ≤ Cf g K } by the following equality
Bg, f = g, B † f . Note that the scalar product appearing in the last definition should be understood as pairing defined by the original scalar product of H. The triplet adjoint operator coincides with the standard adjoint operator in the case K = H = K∗ . Otherwise the triplet adjoint operator B † is different from the adjoint operator B ∗ - operator adjoint to B considered as an operator in the Hilbert space H ⊃ K. Consider the restriction AHn−2 of the operator A to the Hilbert space Hn−2 . This operator is a self-adjoint operator in this Hilbert space with the domain 4 We are going to drop the subindex −n whenever it is possible to make the presentation more transparent.
444
Kurasov
IEOT
Dom (AHn−2 ) = Hn . The triplet adjoint operator A†Hn−2 coincides with the operator AH−n+2 , which the resctriction of A (acting in the scale of Hilbert spaces as described above) to the space H−n+2 . It is also an extension of the original self-adjoint operator A acting in H. The domain of the triplet adjoint operator coincides with the space H−n+4 Dom (A†Hn−2 ) = H−n+4 . Summing up we conclude that the triplet adjoint operator to AHn−2 with respect to the triplet Hn−2 ⊂ H ⊂ H−n+2 coincides with the operator AH−n+2 . We define the minimal operator Amin corresponding to the formal expression (1.1) as the restriction of the operator AHn−2 to the domain of function orthogonal to ϕ Dom (Amin ) = {ψ ∈ Hn : ϕ, ψ = 0}. (3.4) The operator Amin is densely defined, since ϕ ∈ / H−n+1 ⇒ ϕ ∈ / H−n+2 . Then the maximal operator Amax coincides with the triplet adjoint operator to Amin with respect to the triplet Hn−2 ⊂ H ⊂ H−n+2 Amax = A†min .
(3.5)
The following Lemma describes the maximal operator Amax in detail. Lemma 3.1. The maximal operator Amax is defined on the domain Dom (Amax ) = f = f˜ + f1 g1 ∈ H−n+2 , f˜ ∈ H−n+4 , f1 ∈ C .
(3.6)
by the following formula Amax (f˜ + f1 g1 ) = Af˜ − f1 g1 .
(3.7)
Remark 3.2. In the case n = 2 the minimal operator Amin is a symmetric operator in the original Hilbert space having the domain Dom (Amin ) = {ψ ∈ Dom (A) = H2 (A) : ϕ, ψ = 0}. Then the maximal operator Amax coincides with the usual adjoint operator to Amin with the domain given by (3.6). The action of the adjoint operator is given by (3.7). Proof. The domain of the triplet adjoint operator A†min consists of all elements f ∈ H−n+2 such that the sesquilinear form (A + 1)ψ, f = ψ, (A + 1)f can be estimated as follows |ψ, (A + 1)f | ≤ Cf ψ Hn−2 for all ψ ∈ Dom (Amin ), since the operator Amin is a restriction of the operator A. The last estimate holds for all ψ ∈ Hn , ψ, ϕ = 0 if and only if (A + 1)f = fˆ + f1 ϕ, where fˆ ∈ H−n+2 , f1 ∈ C. It follows that the function f possesses representation (3.6).
Krein’s Formula for H−n –perturbations
Vol. 45 (2003)
445
Suppose now that representation (3.6) holds. Then the sesquilinear form can be written as follows (A + 1)ψ, f
= (A + 1)ψ, f˜ + (A + 1)ψ,
1 ϕ A+1
= ψ, (A + 1)f˜ + 0. It follows that (A + 1)† (f˜ + f1 g1 ) = (A + 1)f˜
and hence (3.7) holds.
The operator Amax will be used to define the maximal operator acting in the extended Hilbert space H. Definition 3.3. The maximal operator Amax acting in the Hilbert space H is the restriction of the operator Amax to the Hilbert space H defined by the following equality Amax ρ = ρAmax
(3.8)
on the following domain Dom (Amax ) = {U ∈ H : Amax ρ(U) ∈ Range(ρ)}. The following lemma describes in details the maximal operator Amax . Lemma 3.4. The maximal operator Amax determined by Definition 1 is defined on the domain Dom(Amax ) by the formula
Amax
= {U = (u1 , u2 , . . . , un−2 , Ur + un−1 gn−1 ), u1 , u2 , . . . , un−2 , un−1 ∈ C, Ur ∈ Hn }
u1 u2 ... un−2 Ur + un−1 gn−1
=
u2 − u1 u3 − u2 ... un−1 − un−2 AUr − un−1 gn−1
(3.9)
.
(3.10)
Proof. Consider any vector U = (u1 , u2 , ..., un−2 , U ) from the domain of the operator Amax and let us denote its image by W = (w1 , w2 , . . . , wn−2 , W ). Then equality (3.8) can be written as follows w1 g1 + w2 g2 + . . . + wn−3 gn−3 + wn−2 gn−2 + W =(u2 − u1 )g1 + (u3 − u2 )g2 + . . . + (un−2 − un−3 )gn−3 − un−2 gn−2 + AU. (3.11)
446
Kurasov
IEOT
We conclude that w1 w2 ... wn−3 W + wn−2 gn−2
= =
u2 − u1 ; u3 − u2 ; (3.12)
= un−2 − un−3 ; = AU − un−2 gn−2 .
The last equality can be written as W + U + wn−2 gn−2 + un−2 gn−2 = (A + 1)U and therefore U=
1 1 (W + U ) + (wn−2 + un−2 ) gn−2 . A+1 A+1
It follows that the element U possesses the following representation U = Ur + un−1 gn−1 , where Ur ∈ Hn , un−1 ∈ C. Then equality (3.11) can be written as w1 g1 + w2 g2 + . . . + wn−2 gn−2 + W =
(u2 − u1 )g1 + (u3 − u2 )g2 + . . . + (un−1 − un−2 )gn−2 + AUr − un−1 gn−1 ,
and one can deduce that formula (3.10) holds.
The spectrum of the operator Amax covers the whole complex plane. Indead consider any complex number λ. Then the element 1 1 (1 + λ) (1 + λ) ... ... U= = n−3 n−3 (1 + λ) (1 + λ) n−1 A + 1 (1 + λ) gn−1 (1 + λ)n−2 gn−1 + (1 + λ)n−2 gn−1 A−λ A−λ solves the equation Amax U = λU. Note that the last formula reads as follows in the special case λ = −1 1 1 0 0 = − ... . . . . Amax 0 0 0 0 Let us calculate the adjoint operator Amin = A∗max . Note that the operator Amin is different from the minimal operator Amin considered earlier. In fact the operator Amin is a restriction of the operator Amin .
Vol. 45 (2003)
Krein’s Formula for H−n –perturbations
447
Lemma 3.5. The operator Amin , adjoint to Amax in H, is defined on the domain Dom (Amin ) = {U = (u1 , u2 , . . . , un−2 , Ur ); u1 , u2 , . . . , un−2 ∈ C, Ur ∈ Hn , un−2 = ϕ, Ur } by the formula
Amin
u1 u2 ... un−2 Ur
=
−u1 u1 − u2 ... un−3 − un−2 AUr
(3.13)
.
(3.14)
Proof. Consider arbitrary elements U ∈ Dom (Amax ) and V = (v1 , v2 , . . . , vn−2 , V ) ∈ H. The sesquilinear form of the operator Amax is
(Amax + 1)U, V
=
u ¯ 2 v1 + u ¯ 3 v2 + . . . + u ¯n−1 vn−2 +(A + 1)Ur , (1 + A)n−2 V
=
¯ 3 v2 + . . . + u ¯n−2 vn−3 u ¯ 2 v1 + u
(3.15)
+¯ un−1 vn−2 − gn−1 , (1 + A)n−1 V +Ur + un−1 gn−1 , (1 + A)n−1 V . Consider first the subset of elements U ∈ Dom (Amax ) with uk = 0, k = 1, 2, . . . , n − 1. Then the last term in (3.15) is a bounded functional with respect to U ∈ H if and only if V = Vr ∈ Hn . Consider next arbitrary U ∈ Dom (Amax ). Since the functional U → un−1 is not bounded in the norm of H, the last formula determines bounded linear functional if and only if the expression in { } vanishes, i.e. vn−2 = gn−1 , (1 + A)n−1 Vr ≡ ϕ, Vr .
(3.16)
Hence the domain of Amin is formed by the elements possessing the representation V = (v1 , v2 , . . . , vn−2 , Vr ), Vr ∈ Hn , vk ∈ C and satisfying (3.16). Taking into account these relations, the formula for the adjoint operator can be written as follows
Amax U, V
= U, Amin V =
¯2 (v1 − v2 ) + . . . + u ¯n−2 (vn−3 − vn−2 ) u ¯ 1 v1 + u +Ur + un−1 gn−1 , (1 + A)n−2 AVr .
It follows that the action of the minimal operator is given by (3.14).
448
Kurasov
IEOT
The operator Amin is an extension of the operator Amin . Indead the domain of Amin defined by (3.4) belongs to Dom (Amin ) and moreover Amin = Amin |Dom(Amin ) .5 The domain of the minimal operator Amin is contained in the domain of the maximal operator Amax , but the minimal operator does not coincide with the restriction of the maximal operator to the domain of the minimal one. Therefore no restriction of the operator Amax is self-adjoint like in the case of H−3 perturbations [23]. Therefore no self-adjoint operator can be associated with the formal operator (1.1). This can also be seen from the following lemma. Lemma 3.6. The boundary form of the maximal operator Amax is given by
Amax U, V − U, Amax V =
0 −1 ... 0 0 0
1 0 ... 0 0 0
... 0 ... 0 ... ... ... 0 . . . −1 ... 0
0 0 ... 1 0 −1
0 0 ... 0 1 0
u1 u2 ... un−2 un−3 ϕ, Ur
,
v1 v2 ... vn−2 vn−3 ϕ, Vr
(3.17)
Proof. The following straightforward calculations prove the Lemma
Amax U, V − U, Amax V = −
=
u2 − u1 u3 − u2 ... un−1 − un−2 AUr − un−1 gn−1 u1 u2 ... un−2 Ur + un−1 gn−1
,
,
v1 v2 ... vn−2 Vr + vn−1 gn−1
v2 − v 1 v3 − v 2 ... vn−1 − vn−2 AVr − vn−1 gn−1
+¯ u 2 v1 + u ¯ 3 v2 + . . . + u ¯n−1 vn−2 − u ¯ 1 v2 − u ¯ 2 v3 − . . . − u ¯n−2 vn−1 +Ur , ϕvn−1 − u ¯n−1 ϕ, Vr .
We have used that ϕ = (1 + A)n−1 gn−1 in these calculations. 5 The
operators Amin and Amin coincide only in the cases n = 1, 2.
Vol. 45 (2003)
Krein’s Formula for H−n –perturbations
The matrix describing the boundary form 0 −1 0 . . . 0 0 1 0 −1 . . . 0 0 0 1 0 ... 0 0 B≡ ... ... ... ... ... ... 0 0 0 . . . 0 −1 0 0 0 ... 1 0 0 0 0 ... 0 1
0 0 0 ... 0 −1 0
449
is symplectic and has rank n for even n and n − 1 for odd n. Hence any symmetric restriction of the operator Amax is described by at least [ n2 ] boundary conditions.6 Such restriction cannot be self-adjoint, since the kernel of the operator Amax − λ for nonreal λ has dimension one. We have proven for the second time that no restriction of the operator Amax is self-adjoint in the Hilbert space H and no self-adjoint operator corresponds to formal expression (1.1) in the case ϕ ∈ H−n \ H−n+1 , n ≥ 4. To define a non self-adjoint operator corresponding to this formal expression the class of regular operators will be introduced in the following section. If n = 3, then the rank of the matrix B is 2 and all Lagrangian planes of the boundary form are described by one condition. Thus the restrictions of Amax to the corresponding subspaces are self-adjoint operators [23]. One can proceed now along two possible lines: 1. Construct a non self-adjoint operator corresponding to (1.1). 2. Consider the maximal common symmetric restriction of the operators Amin and Amax and describe all its self-adjoint extensions. We decided to follow the first possibility, since the resolvent of the operator obtained in this way is given by formula (4.15) similar to Krein’s formula for generalized resolvents. The second approach is described in [25] (for n = 4 only).
4. Regular Operators and Supersingular Perturbations of Self-adjoint Operators The operator corresponding to the formal expression (1.1) is a certain restriction of the maximal operator. In this section we are going to study the set of regular restrictions of the maximal operator. All regular restrictions of the maximal operator are characterized by the following theorem. Theorem 4.1. A restriction A of Amax is a regular operator if and only if there exist real numbers a, b, not equal to zero simultaneously, such that Dom A is the restriction of Dom Amax by aϕ, Ur + bun−1 − aun−2 = 0. 6 [·]
denotes the integer part here.
(4.1)
450
Kurasov
IEOT
Proof. The domain Dom (Amax ) of the maximal operator contains the domain Dom (Amin ) of its adjoint. Therefore every regular restriction of Amax is an extension of Amin . 7 The domain Dom (Amax ) consists of all elements U ∈ H possessing the representation U = (u1 , u2 , . . . , un−2 , Ur + un−1 gn−1 ), where Ur ∈ Hn , uk ∈ C, k = 1, 2, . . . , n − 1. The domain Dom (Amin ) of the adjoint operator is a subdomain of Dom (Amax ) characterized by the boundary conditions un−1 = 0, ϕ, Ur − un−2 = 0. Thus the dimension of the quotient space Dom (Amax )/Dom (Amin ) is equal to 2. Any linear subset D of Dom (Amax ) which does not coincide with Dom (Amin ) and Dom (Amax ) is described by the boundary conditions of the form a (ϕ, Ur − un−2 ) + bun−1 = 0,
(4.2)
where (a, b) is a two dimensional nonzero complex vector. (If both parameters a and b are equal to zero, then the boundary condition (4.2) is satisfied by all functions from Dom (Amax ) and the linear subset coincides with the domain of maximal operator.) The foregoing shows that a regular restriction has a domain of the form (4.2). It has to be determined under what necessary and sufficient conditions on a and b a restriction of the maximal operator to a domain of the form (4.2) is regular. The sesquilinear form of the operator Amax |D is given again by formula (3.15), where now U ∈ D. Consider vectors U with un−2 = un−1 = ϕ, Ur = 0. Then the scalar product Ur , (A + 1)n−1 V generates a bounded linear functional with respect to the vector (0, 0, . . . , 0, Ur ) ∈ H and the standard norm in H if and only if the following representation holds8 (A + 1)n−1 V = cϕ + f˜, where c ∈ C, f˜ ∈ H−n+2 . This implies that V = cgn−1 +
1 f˜, (A + 1)n−1
and it follows that the vector V possesses the representation V = Vr + vn−1 gn−1 , 7 This
procedure is similar to the one used in the extension theory of symmetric operators. that Ur is orthogonal to ϕ.
8 Remember
Vol. 45 (2003)
Krein’s Formula for H−n –perturbations
451
where Vr ∈ Hn , vn−1 ∈ C. Then the sesquilinear form is given by
(A + 1)U, V = Ur + un−1 gn−1 , (A + 1)n−1 Vr − u ¯n−1 ϕ, Vr + vn−1 Ur , ϕ ¯n−2 vn−3 + . . . + u ¯ 3 v2 + u ¯ 2 v1 . +¯ un−1 vn−2 + u The domain of the operator adjoint to Amax |D is characterized by the condition that the last formula determines bounded a certain linear functional with respect to U ∈ H. We are going consider all possible values of the parameters a and b and study the question whether the domain of the adjoint operator coincides with the domain of the restricted operator or not. Let us call admissible the two dimensional vectors leading to regular restrictions of the maximal operator. The following two cases cover all possible values of the the parameters. Case 1. General: a = 0, b arbitrary. The boundary condition can be presented in the form b ϕ, Ur − un−2 = − un−1 a and the sesquilinear form of the operator is given by
(A + 1)U, V ¯b ¯n−1 −ϕ, Vr − vn−1 + vn−2 = Ur + un−1 gn−1 , (A + 1)n−1 Vr + u a ¯ +¯ un−2 (vn−3 + vn−1 ) + u ¯n−3 vn−4 + . . . + u ¯ 3 v2 + u ¯ 2 v1 . The last expression determines a bounded linear functional if and only if the following relation holds a ¯ (ϕ, Vr − vn−2 ) + ¯bvn−1 = 0. This condition coincides with (4.2) if and only if the complex numbers a and b have the same phase. Hence without loss of generality the constants a, b can be chosen real. Any vector (a, b) ∈ R2 is admissible. Case 2. Special a = 0, b = 0. The boundary condition takes the form un−1 = 0.
(4.3)
452
Kurasov
IEOT
Hence the sesquilinear form of the operator is given by
(A + 1)U, V ¯n−2 vn−3 = Ur + un−1 gn−1 , (A + 1)n−1 Vr + u +Ur , ϕvn−1 + u ¯n−3 vn−4 + . . . + u ¯ 3 v2 + u ¯ 2 v1 . This form defines bounded linear functional with respect to U ∈ H if and only if vn−1 = 0, since Ur , ϕ is not a bounded linear functional. The last condition coincides with (4.3). Without loss of generality the constant b can be chosen real. Every vector (0, b), b = 0, is admissible. Summing up our studies we conclude that the set of regular restrictions of the operator Amax can be characterized by the boundary conditions (4.1) with real a and b not equal to zero simultaneously. The theorem states that all regular restrictions of the operator Amax are described by nontrivial real two dimensional vectors (a, b). The length of the vector (a, b) plays no rˆ ole and therefore all boundary conditions can be parametrized by one real parameter - ”angle” θ ∈ [0, π) as follows: sin θϕ, Ur + cos θun−1 − sin θun−2 = 0.
(4.4)
The following definition will be used. Definition 4.2. The operator Aθ is the restriction of the maximal operator Amax to the set of functions satisfying boundary conditions (4.4). The domain of the operator Aθ is formed by the functions from Dom (Amax ) (given by (3.9)) subject to the boundary conditions (4.4). The action of the operator Aθ is given by (3.10). Thus the regular operator corresponding to the formal expression (1.1) is not defined uniquely. Like in the case of H−2 and H−3 -perturbations a one parameter family of operators has been constructed. Let us calculate the operator adjoint to Aθ . The domain of this operator coincides with the domain Dom (Aθ ). The sesquilinear form of the operator Aθ can be presented by the following expression using the fact, that the functions from the domains of the operators Aθ and A∗θ satisfy (4.4)
(Aθ + 1)U, V = Ur + un−1 gn−1 , (A + 1)n−1 Vr + u ¯n−2 (vn−1 + vn−3 ) +
n−4 k=1
u ¯k+1 vk ,
Krein’s Formula for H−n –perturbations
Vol. 45 (2003)
and it follows that
∗ (Aθ + 1)
v1 v2 ... vn−3 vn−2 Vr + vn−1 gn−1
=
453
0 v1 ... vn−4 vn−1 + vn−3 (A + 1)Vr
.
Hence the action of the operator A∗θ is given by −v1 v1 v v − v2 2 1 ... . .. ∗ = Aθ − vn−3 v v n−3 n−4 vn−1 + vn−3 − vn−2 vn−2 Vr + vn−1 gn−1 AVr − vn−1 gn−1
.
(4.5)
The real and imaginary parts of the operator Aθ are given by
Aθ = Aθ + iAθ ;
u1 u2 ... = (Aθ ) un−3 un−2 Ur + un−1 gn−1 0 −i 0 i 0 −i 0 i 0 1 . . . . . . . .. Aθ = 2 0 0 0 0 0 0 0 0 0
1 2 u2 − u1 1 (u + u1 ) − 3 2
...
u2
1 2 (un−2 + un−4 ) − un−3 un−1 + 12 un−3 − un−2
... ... ... ... ... ... ...
AUr − un−1 gn−1 0 0 0 0 0 0 0 0 0 ... ... ... . 0 −i 0 i 0 0 0 0 0
;
(4.6)
The real part of the operator Aθ is a self-adjoint operator on the domain Dom (Aθ ). The imaginary part of Aθ is a bounded self-adjoint operator, which does not depend on the parameter θ. Let us study the operator A0 in more details. This operator is equal to the orthogonal sum of two operators acting in the spaces Cn−2 and Hn−2 . Indead the domain of the operator A0 can be decomposed as follows Dom (A0 ) = Cn−2 ⊕ Hn ⊂ Cn−2 ⊕ Hn−2 ≡ H. The two operators appearing in the corresponding decomposition of the operator A0 A0 = T ⊕ A,
454
Kurasov
are the operator in Cn−2 given by the upper −1 1 0 0 0 −1 1 0 0 0 −1 1 0 0 0 −1 T= ... ... ... ... 0 0 0 0 0 0 0 0
IEOT
triangular matrix ... 0 0 ... 0 0 ... 0 0 ... 0 0 ... ... ... . . . −1 1 . . . 0 −1
and the operator A in Hn−2 with the domain Hn . The resolvent of the operator A0 for arbitrary nonreal λ can easily be calculated −1 −1 −1 −1 . . . (1+λ) n−2 1+λ (1+λ)2 (1+λ)3 0 −1 −1 −1 . . . 2 n−3 (1+λ) (1+λ) (1+λ) 1 1 −1 −1 ⊕ = 0 . (4.7) 0 . . . (1+λ) (1+λ)n−4 A−λ A0 − λ ... ... ... ... ... −1 0 0 0 ... 1+λ Let us study now the spectrum of the operator Aθ . The following theorem implies that the spectrum is real, since the resolvent of Aθ wxists and is a bounded operator for nonreal values of the spectral parameter. Theorem 4.3. The resolvent of the operator Aθ for all nonreal λ is given by the (n − 1) × (n − 1) bounded matrix operator 1 1 = Aθ − λ A0 − λ 0 0 ... 0 0 ... sin θ ... ... ... − D(λ) 0 0 ... 0 0 ...
0
1 (1+λ)n−1
1 1 (1+λ)n−2 A−λ ϕ, ·
0
1 (1+λ)n−2
1 1 (1+λ)n−3 A−λ ϕ, ·
...
...
...
0
1 (1+λ)2
1 1 1+λ A−λ ϕ, ·
0
1 1 1+λ A−λ gn−2
1 A−λ gn−2
1 A−λ ϕ, ·
where the function D(λ, θ) is the following Nevanlinna function 1+λ 1 D(λ, θ) = ϕ, gn−1 − sin θ + cos θ. A−λ 1+λ
(4.8)
(4.9)
Vol. 45 (2003)
Krein’s Formula for H−n –perturbations
455
Proof. Consider arbitrary F = (f1 , f2 , . . . , fn−2 , F ) ∈ H. Then the resolvent equation f1 u1 f2 u2 = ... ... (4.10) (A − λ) fn−2 un−2 Ur + un−1 gn−1 F together with the boundary condition (4.4) imply that −(1 + λ) 1 0 ... 0 0 0 0 −(1 + λ) 1 ... 0 0 0 0 0 −(1 + λ) . . . 0 0 0 ... ... ... ... ... ... ... 0 0 0 . . . −(1 + λ) 1 0 1+λ 0 0 0 ... 0 −ϕ, gn−1 1 A−λ 0 0 0 ... − sin θ cos θ sin θ ×
u1 u2 u3 ... un−2 un−1 ϕ, Ur
=
f1 f2 f3 ... fn−2 1 ϕ, A−λ F 0
.
(4.11)
To derive the last equation we used the following transformation of the last equation in the system (4.10) (A − λ)Ur − (1 + λ)un−1 gn−1 = F ⇒ ϕ, Ur − (1 + λ)un−1 ϕ,
(4.12)
1 1 gn−1 = ϕ, F . A−λ A−λ
The determinant of the matrix appearing in the last equation is equal to (−1)n−1 (1 + λ)n−2 D(λ, θ) and it vanishes for nonreal λ only if D(λ, θ) = 0. The 1+λ 1 1 imaginary part of the function ϕ, is given by ϕ − A − λ (A + 1)n−1 1+λ (A+1)2 1 1+λ 1 1 1 ϕ, A−λ , ϕ − = y ϕ, ϕ + n−1 2 2 n 2 2 (A+1) 1+λ (A−x) +y (A+1) (1+x) +y where λ = x + iy, x, y ∈ R. The imaginary part cannot vanish for nonreal values of λ if θ = 0. In the case θ = 0 the function D(λ, 0) ≡ 1 is constant. We conclude that the linear system (4.11) has unique solution for all nonreal λ. It follows that the spectrum of the operator Aθ is real.
456
Kurasov
IEOT
To calculate the resolvent exactly consider the system of equations for un−2 , un−1 , ϕ, Ur
1 0 fn−2 un−2 1+λ 1 F gn−1 1 un−1 = ϕ, A−λ 0 −ϕ, A−λ ϕ, U 0 r − sin θ cos θ sin θ (4.13) The solution to this linear system reads as follows −(1 + λ)
un−2
= −
un−1
= −
ϕ, Ur = −
1+λ 1 (sin θϕ, A−λ gn−1 + cos θ)fn−2 + sin θϕ, A−λ F
(1 + λ)D(λ, θ) 1 (1 + λ)ϕ, A−λ F + fn−2 sin θ (1 + λ)D(λ, θ)
;
;
1+λ 1 sin θϕ, A−λ gn−1 fn−2 + (sin θ − (1 + λ) cos θ) ϕ, A−λ F
(1 + λ)D(λ, θ)
.
(4.14) Then all other components of the vector u can be calculated from the recursive relations 1 1 ul+1 − fl , l = 1, 2, . . . , n − 3, ul = 1+λ 1+λ which coincide with the first n − 3 equations of the system (4.11). The following formula holds n−3
1 1 ul = un−2 − fm . n−2−l (1 + λ) (1 + λ)m+1−l m=l
The component U can be calculated from (4.12) U
= Ur + un−1 gn−1 1 1 F + (1 + λ)un−1 gn−1 + un−1 gn−1 = A−λ A−λ 1 1 F + un−1 gn−2 . = A−λ A−λ
This completes the calculation of the resolvent of the operator Aθ given by formula (4.8) for all nonreal λ. The theorem implies that the spectrum of the operator Aθ is real. Consider the restriction of the resolvent to the subspace Hn−2 ⊂ H combined with the
Krein’s Formula for H−n –perturbations
Vol. 45 (2003)
457
embedding ρ 1 1 |H = ρ Aθ − λ n−2 A−λ −
1 1+λ 1 n−2 cot θ + ϕ, A−λ (λ + 1) (A+1)n−1 ϕ −
1 1+λ
1 1 ϕ, ϕ, · ¯ A−λ A−λ
(4.15) The last formula is analogous to Krein’s formula connecting the resolvents of two self-adjoint extensions of one symmetric operator and is very similar to formula (2.6) describing the restricted resolvent of the self-adjoint operator corresponding to the singular H−3 -perturbations. The main difference between this formula and the well-known Krein’s formula is that the conventional Krein’s formula describes the resolvent of the self-adjoint operator, while the formula obtained comes from a certain non self-adjoint operator if the perturbation is singular enough ϕ ∈ H−n , n ≥ 4. The last formula can be called Krein’s formula for supersingular interactions. The spectral properties of the operator are described by the generalized Nevanlinna function 1 1 1+λ n−2 ϕ − Q(λ) = (λ + 1) cot θ + ϕ, . (4.16) A − λ (A + 1)n−1 1+λ The zeroes of this function determine the singularities of the resolvent. The func1 1+λ 1 is a standard Nevanlinna function tion cot θ + ϕ, ϕ − A − λ (A + 1)n−1 1+λ tending to −∞ and +∞ when λ → −∞ and λ → −1− respectively. Therefore the function has at least one zero in the interval (−∞, −1). Another one zero can be situated in the interval (−1, 0) depending on the behavior of the function 1 1+λ ϕ, A−λ (A+1)n−1 ϕ at the origin and the coupling parameter θ. Let λ0 < 0, λ0 = −1 be a zero of the function Q(λ). Then the vector 1
(1+λ0 )n−2 1 (1+λ0 )n−3
...
1 1+λ0 1 1 A−λ0 (A+1)n−2 ϕ
(4.17)
is an eigenvector of the operator Aθ corresponding to the eigenvalue λ0 . The point λ = −1 is an eigenvalue of the operator Aθ with the eigenvector (1, 0, 0, 0, . . . , 0, 0) . n−1
1 (λ+1) The function Q−n (λ) = ϕ, A−λ (A+1)n−1 ϕ , consituting the nontrivial part of the generalized Nevanlinna function Q(λ) appearing in (4.16), is an n − 1-times regularized resolvent function
Q−n (λ)
formally
=
ϕ,
1 λ+1 1 (1 + λ)n−2 ϕ−ϕ, ϕ−ϕ, ϕ−. . .−ϕ, ϕ. A−λ A+1 (A + 1)2 (A + 1)n−1
458
Kurasov
IEOT
5. Conclusions Rank one singular perturbations of self-adjoint determined by arbitrary vectors from the class H−n have been defined in this article. It has been shown that such operators can be defined in the class of non self-adjoint operators acting in a certain extended Hilbert space. The final operator obtained is nevertheless close to a self-adjoint one - the imaginary part of the operator is a bounded operator. It has been proven that the spectrum of the perturbed operator is pure real. It remains to study in more details the spectral properties of the operator obtained. It is not clear whether the operator constructed is similar to a certain self-adjoint one. These questions will be considered in one of the future publications. The approach developed in this paper has to be generalized in order to include perturbations of not finite rank following the main ideas of [26]. Acknowledgment The author would like to thank The Swedish Royal Academy of Sciences for financial support. He is grateful to S.Albeverio, A.Dijksma, H.Langer, S.Naboko, B.Pavlov, Yu.Shondin, and K.Watanabe, for many fruitful discussions.
References [1] S. Albeverio, F. Gesztesy, R. Hoegh-Krohn, and H. Holden, ”Solvable models in quantum mechanics”, Springer-Verlag, 1988. [2] S.Albeverio and P.Kurasov, Rank one perturbations, approximations, and self-adjoint extensions, J. Funct. Anal., 148, 1997, 152–169. [3] S.Albeverio and P.Kurasov, Rank one perturbations of not semibounded operators, Int. Eq. Oper. Theory, 27, 1997, 379–400. [4] S.Albeverio and P.Kurasov, Singular perturbations of differential operators, Cambridge Univ. Press, 2000 (London Mathematical Society Lecture Notes 271). [5] I.Andronov, Zero-range potential model of a protruding stiffener, J.Phys.A, 32 , 1999, L231–238. [6] J.Arazy and L.Zelenko, Finite-dimensional perturbations of self-adjoint operators, Int. Eq. Oper. Theory, 34 (1999), 127–164. [7] F.A.Berezin and L.D.Faddeev, Remark on the Schr¨ odinger equation with singular potential, Dokl. Akad. Nauk SSSR, 137, 1011–1014, 1961. [8] J.Boman and P.Kurasov, Finite rank singular perturbations and distributions with discontinuous test functions, Proc. AMS 126 (1998), 1673–1683. [9] Yu.N.Demkov and V.N.Ostrovsky, Zero-range potentials and their applications in atomic physics, Plenum Press, New York and London, 1988. [10] Yu.N.Demkov and G.F.Drukarev, Loosely bound particle with nonzero orbital angular momentum in an electric or magnetic field, Sov. Phys. JETP, 54 (4), 1981, 650–656. [11] Yu.N.Demkov and G.F.Drukarev, Potential curves for the p-state of the system AB − in the approximation of small radius potentials, in: Problems of atomic collisions, vyp. 3, Leningrad Univ. Press, Leningrad, 1986, 184–195.
Vol. 45 (2003)
Krein’s Formula for H−n –perturbations
459
[12] J.F.van Diejen and A.Tip, Scattering from generalized point interactions using self– adjoint extensions in Pontryagin spaces, J. Math. Phys., 32, 1991, 630–641. [13] A.Dijksma, H.Langer, Yu.Shondin, and C.Zeinstra, Self–adjoint operators with inner singularities and Pontryagin spaces, Operator Theory: Advances and Applications, 118 (2000), 105–175. [14] F.Gesztesy, K.Makarov, and Tsekanovsky, An addendum to Krein’s formula, J. Math. Anal. Appl., 222 (1998), 594–606. [15] F.Gesztesy and B.Simon, Rank-one perturbations at infinite coupling, J. Funct. Anal., 128, 1995, 245–252. [16] S.Hassi and H. de Snoo, One-dimensional graph perturbations of self–adjoint relations, Ann. Acad. Sci. Fenn. Math., 22, 1997, 123–164. [17] Yu.E.Karpeshina, Zero-Range Model of p-scattering by a Potential Well, Preprint Forschungsinstitut f¨ ur Mathematik, ETH, Z¨ urich, 1992. [18] A.Kiselev and I.Popov, An indefinite metric and scattering by regions with a small aperture, Mat. Zametki, 58, 1995, 837–850, 959. [19] A.Kiselev and B.Simon, Rank one perturbations with infinitesimal coupling, J. Funct. Anal., 130, 1995, 345–356. [20] V.Koshmanenko, Singular Quadratic Forms in Perturbation Theory, Kluwer, Dordrecht, 1999. [21] M.Krein, On Hermitian operators whose deficiency indices are 1, Comptes Rendus (Doklady) Acad. Sci. URSS (N.S.), 43, 131-134, 1944. [22] P.Kurasov and S.T.Kuroda, Krein’s formula in perturbation theory, in preparation. [23] P.Kurasov and K.Watanabe, On rank one H−3 -perturbations of positive self-adjoint operators, In: F.Gesztesy, H.Holden, J.Jost, S.Paycha, M.R¨ ockner, S.Scarlatti (Eds.), Stochastic Processes, Physics and Geometry: New Interplays, II, CMS conference proceedings, 29, AMS, Providence, 2000, pp. 413–422. [24] P.Kurasov and K.Watanabe, On H−4 -perturbations of self-adjoint operators, Operator Theory:Advances and Applications, 126 (2001), 179–196. [25] P.Kurasov and K.Watanabe, Supersingular H−4 -perturbations of self-adjoint operators, in preparation. [26] S.T.Kuroda and H.Nagatani, H−2 -construction and some applications, in: Mathematical Results in Quantum Mechanics (QMath 7, Prague, 1998), J.Dittrich, P.Exner, M.Tartar (eds.), Operator Theory: Advances and Applications, 108 Birkh¨ auser Verlag, 1999. [27] M.Naimark, On spectral functions of a symmetric operator, Bull. Acad. Sciences URSS, 7, 285-296, 1943. [28] B.Pavlov, The theory of extensions and explicitly solvable models, Uspekhi Mat. Nauk, 42, 99-131, 1987. [29] B.Pavlov, Boundary conditions on thin manifolds and the semiboundedness of the three-body Schr¨ odinger operator with point potential, Mat. Sb. (N.S.), 136, 163-177, 1988. [30] B.S.Pavlov and I.Popov, Scattering by resonators with small and point holes, Vestnik Leningrad. Univ. Mat. Mekh. Astronom., vyp. 3, 1984, 116–118.
460
Kurasov
IEOT
[31] B.S.Pavlov and I.Popov, An acoustic model of zero-width slits and the hydrodynamic stability of a boundary layer, Teoret. Mat. Fiz., 86, 1991, 391–401. [32] I.Popov, The Helmholtz resonator and operator extension theory in a space with indefinite metric, Mat. Sb., 183, 1992, 3–27. [33] M.Reed and B.Simon, Methods of modern mathematical physics II, Academic Press, New York, 1972. [34] S.N.Saakjan, Theory of resolvents of a symmetric operator with infinite defect numbers, Akad. Nauk. Armjan. SSR Dokl., 41, 193-198, 1965. [35] Yu.Shondin, Perturbation of differential operators on high-codimension manifold and the extension theory for symmetric linear relations in an indefinite metric space, Teoret. Mat. Fiz., 92, 1992, 466–472. [36] Yu.Shondin, Quantum mechanical models in Rn connected with extensions of the energy operator in a Pontryagin space, Teoret. Mat. Fiz., 74, 1988, 331–344. [37] B.Simon, Spectral analysis of rank one perturbations and applications, in: Mathematical quantum theory. II. Schr¨ odinger operators (Vancouver, BC, 1993), 109–149, CRM Proc. Lecture Notes, 8, AMS, Providence, RI, 1995. Pavel Kurasov Dept. of Mathematics, Lund Institute of Technology, Box 118, 221 00 Lund, Sweden E-mail:
[email protected] Submitted: February 25, 2001 Revised: March 22, 2002
To access this journal online: http://www.birkhauser.ch